CLIMA Score Methodology

How We Calculate Climate Risk for Mortgage Underwriting

Download This Document

Save as PDF to share with investors and stakeholders

Back to CLIMA Demo

Overall Formula

Final CLIMA Score = (Factor_A × 0.30) +
(Factor_B × 0.20) +
(Factor_C × 0.20) +
(Factor_D × 0.15) +
(Factor_E × 0.15)

Key Principle: Higher score = Safer (lower climate risk). All scores normalized to 0-10 scale.

Factor A: Climate Risk Progression 30% Weight

The most important factor - combines historical disaster frequency with future climate projections.

Historical Component (15% of Factor A)

Analyzes 20-year disaster trends using FEMA disaster declarations:

# 20-year disaster trend analysis
disaster_count_last_20y = count_disasters_in_last_20_years(county_fips)
disaster_count_prior_20y = count_disasters_in_prior_20_years(county_fips)

# Calculate trend ratio (ratio of 1.0 = no change)
trend = (disaster_count_last_20y + 1) / (disaster_count_prior_20y + 1)

# Scale trend to 0-10 (higher trend = lower score = more risk)
A_hist = max(0, min(10, 10 - ((trend - 0.5) / (3.0 - 0.5)) * 10))
            

Projected Component (85% of Factor A)

Uses real climate projections from NOAA/NASA for 2050 using SSP2-4.5 and SSP5-8.5 scenarios:

Peril Type	Weight	Data Source	Projection
Flood Risk	40%	NOAA Sea Level Rise	% land area at risk of inundation
Wildfire Risk	30%	USFS Wildfire Hazard Potential	WHP index change 2020–2050
Heat Risk	20%	NASA NEX-GDDP	Days ≥95°F temperature change
Wind/Hurricane	10%	NOAA Climate Models	Category 3+ hurricane frequency

# Get climate projection deltas for 4 perils
deltas = get_all_projection_deltas(county_fips, scenario="SSP2-4.5")

# Convert each peril delta to 0-10 score
# delta range: [-1, +1] where +1 = large increase in risk
for peril, delta in deltas.items():
    peril_score = max(0, min(10, 5 - (delta * 5)))

# Weighted average across all perils
A_proj = weighted_average(peril_scores, peril_weights)
            

Why 15% Historical / 85% Projected? Mortgages are 30-year commitments. Climate change is accelerating, making future projections more relevant than historical patterns.

Factor B: Relief Allocations per Capita 20% Weight

Measures how much federal disaster aid the area receives - proxy for underlying risk level.

# Get FEMA Public Assistance data (20 years)
pa_total = get_public_assistance_funding(county_fips, years_back=20)

# Calculate per capita assistance
per_capita = pa_total / 100000  # Assumes ~100K county population

# Scale to 0-10 (more aid = higher underlying risk = lower score)
B_score = min(per_capita / 1000, 10)  # $1000 per capita = maximum score
        

Logic: Areas receiving more federal disaster aid typically face higher underlying risk or have weaker local resilience.

Factor C: Municipal Fiscal Stress 20% Weight

Measures local government's ability to respond to disasters using multiple data sources:

Component	Weight in Factor C	Data Source	Scoring Method
Unemployment	50%	BLS Local Area Unemployment Statistics	3-year average, bracket scoring (≤3% = 10.0, ≤4% = 8.5, etc.)
Bond Market Health	25%	FRED Municipal Bond Spreads	State-level municipal bond yield spread vs. Treasury
Government Finances	25%	Census Government Finances	County debt-to-revenue ratio

# Calculate composite fiscal stress score
unemp_score = bracket_score(unemployment_3yr_avg)  # 1.0-10.0 based on brackets
bond_score = percentile_scale(municipal_spread, invert=True)
gov_score = percentile_scale(debt_ratio, invert=True)

# Weighted average
C_score = (unemp_score × 0.50) + (bond_score × 0.25) + (gov_score × 0.25)
        

Factor D: Personal Finance Resilience 15% Weight

Measures household ability to weather disasters using Census ACS 5-year data:

# Get Census ACS 5-year data
income = get_median_household_income(county_fips)        # e.g., $65,000
home_value = get_median_home_value(county_fips)          # e.g., $350,000
housing_burden = get_housing_cost_burden(county_fips)    # e.g., 28%

# Calculate Value-to-Income ratio (mortgage affordability)
vti_ratio = home_value / (income × 1.0)

# Scale each component to 0-10
income_score = min(10, max(0, (income - 30000) / 50000 * 10))
vti_score = max(0, min(10, 10 - (vti_ratio - 2) / 3 * 10))
burden_score = max(0, min(10, 10 - housing_burden / 5))

# Weighted average (higher income, lower VTI, lower burden = better)
D_score = (income_score × 0.4) + (vti_score × 0.3) + (burden_score × 0.3)
        

Components:

Income Score (40%): Median household income, scaled $30K-$80K range
Value-to-Income (30%): Home value relative to income (affordability)
Housing Burden (30%): % of income spent on housing costs

Factor E: Insurance Availability 15% Weight

Measures insurance market functionality using NFIP (National Flood Insurance Program) data:

# Get NFIP data for the county
policies_per_unit = get_nfip_policies_per_housing_unit(county_fips)
avg_premium = get_nfip_average_premium(county_fips)
claim_frequency = get_nfip_claims_per_policy_year(county_fips)

# Scale each component
coverage_score = percentile_scale(policies_per_unit, invert=False)    # Higher = better
premium_score = percentile_scale(avg_premium, invert=True)            # Lower = better
claims_score = percentile_scale(claim_frequency, invert=True)         # Lower = better

# Simple average of available metrics
E_score = (coverage_score + premium_score + claims_score) / 3
        

Metrics:

Coverage Rate: NFIP policies per housing unit (higher = better)
Affordability: Average premium (lower = better)
Claim Frequency: Claims per policy-year (lower = better)

Score Normalization & Scaling

All individual scores are normalized to 0-10 scale using percentile scaling:

def percentile_scale(value, invert=False, fallback_min=None, fallback_max=None):
    """
    Scale any value to 0-10 range
    
    Args:
        value: Raw value to scale
        invert: If True, return 10 - scaled_value (for risk metrics)
        fallback_min/max: Baseline ranges for scaling
    """
    # Use fallback ranges or determine from value magnitude
    min_val, max_val = determine_scaling_range(value)
    
    # Linear scaling to 0-10
    scaled = max(0, min(10, (value - min_val) / (max_val - min_val) * 10))
    
    # Invert if needed (for risk factors where higher = worse)
    if invert:
        scaled = 10 - scaled
    
    return scaled
        

Inversion Logic: For risk factors (like disaster frequency), higher values = worse outcomes = lower scores. The system automatically inverts these scales.

Score Interpretation

Score Range	Grade	Risk Level	Description
9.0-10.0	A+	Minimal Risk	Excellent climate resilience
7.0-8.9	A	Low Risk	Low climate-linked mortgage risk
5.0-6.9	B	Moderate Risk	Standard underwriting appropriate
3.0-4.9	C	Elevated Risk	Consider pricing adjustments
0.0-2.9	D/F	High Risk	Very high risk - consider declining

Configuration & Weights

All weights and parameters are configurable in config.yaml:

scoring:
  weights:
    climate_risk: 0.30           # Factor A: Climate Risk Progression
    relief_allocations: 0.20     # Factor B: Relief Allocation Efficiency  
    fiscal_stress: 0.20          # Factor C: Municipal Fiscal Stress
    personal_finance: 0.15       # Factor D: Personal Finance Resilience
    insurance_coverage: 0.15     # Factor E: Insurance Availability
  
  factorA:
    historical_weight: 0.15      # Historical disasters (15%)
    projected_weight: 0.85       # Climate projections (85%)
    peril_weights:
      flood: 0.40               # Flood risk weight
      wildfire: 0.30            # Wildfire risk weight
      heat: 0.20                # Heat risk weight
      wind: 0.10                # Wind/Hurricane risk weight
        

Key Design Principles

Higher score = safer - inversion applied to risk factors for intuitive interpretation
Public data only - all sources are free and reproducible (FEMA, Census, BLS, NOAA)
Future-focused - 85% weight on climate projections for 30-year mortgage commitments
No look-ahead bias - validation uses only past data at each time point
Transparent weights - no black-box coefficients, all documented and configurable
Industry-exceeding validation - 0.841 default AUC and 0.904 disaster AUC on 145,452 observations and 30,131 disaster county-years

Ready to See CLIMA in Action?

Try the live demo with any ZIP code

Live CLIMA Demo

About CLIMA: CLIMA (Climate-Linked Infrastructure & Mortgage Analytics) is an enterprise-grade climate risk scoring system for mortgage underwriting. Validated on 145,452 observations and 30,131 disaster county-years with 0.841 default prediction AUC and 0.904 disaster prediction AUC, both exceeding industry standard (0.80).

Documentation: Complete methodology, validation reports, and source code are publicly available for regulatory compliance and independent verification.